When working with datasets, the Datastore backend will normalize any empty datasets to the dataset of the calling application. This should be preferred to prepending "s~" to dataset ids as that is not always a valid operation.